1. The tech world goes wild over DeepSeek

DeepSeek is an AI startup specializing in large language models (LLMs), founded in 2023 in Hangzhou, China. Despite entering the market later than major players like OpenAI and Google, DeepSeek has quickly established its position on the global AI map.

The DeepSeek-R1 model, launched in January 2025, made a strong impression with its superior reasoning capabilities while costing less than $6 million to train—an extremely low figure compared to the billions spent by its competitors. This breakthrough not only stirred excitement in the tech community but also sent shockwaves through the financial market: Nvidia’s stock dropped by 17%, wiping out nearly $600 billion in market capitalization—a record decline in U.S. corporate history.

According to Reuters, following the success of DeepSeek-R1, the company is accelerating the development of DeepSeek-R2, expected to launch before May 2025, with enhanced reasoning, programming capabilities, and multilingual support beyond English.

The rise of DeepSeek poses significant challenges for the U.S., forcing policymakers to reassess AI chip export controls and strategies for maintaining technological leadership.

2. Elon Musk introduces Grok-3 – "The world's smartest AI chatbot"

On February 18, Elon Musk’s company, xAI, officially launched Grok-3, its latest AI chatbot, aiming to compete with OpenAI’s ChatGPT and China’s DeepSeek.

Grok-3 boasts ten times the computational power of Grok-2, supported by the Colossus supercomputer cluster with approximately 200,000 GPUs. The model features two distinct thinking modes: "Think" mode for step-by-step analysis, "Big Brain" mode for complex reasoning tasks

Another standout feature is DeepSearch, which allows users to quickly search and synthesize information from the internet and X (formerly Twitter), significantly reducing research time. However, Grok-3's development has sparked controversy. Canada’s privacy watchdog has launched an investigation into whether X used personal data for AI training, raising concerns about user privacy.

Additionally, voice interaction capabilities are being integrated into Grok-3, with the feature expected to roll out in the coming weeks.

The release of Grok-3 marks a significant step for xAI in the global AI race, where major players like OpenAI, Google, and DeepSeek continue to innovate to stay ahead.

3. OpenAI unveils GPT-4.5 – Its largest AI model yet

Recently, OpenAI introduced GPT-4.5, its most advanced AI model, representing a major leap in natural language processing. The model offers: deeper contextual understanding, more natural interactions and reduced "hallucination" errors.

Despite these upgrades, GPT-4.5 still faces limitations inherent in traditional AI training methods, which rely on increasing computational power and expanding training data. OpenAI acknowledged that this approach is reaching its limits, as rising computational costs no longer yield groundbreaking improvements.

As a result, OpenAI is shifting its focus toward Reasoning AI, a new approach expected to overcome existing AI limitations. Unlike current models that rely on statistical predictions, Reasoning AI can engage in multi-step thinking, logical reasoning, and effectively solve complex problems.

OpenAI is currently developing GPT-5, which is anticipated to merge traditional GPT models with Orion (o-series), an advanced Reasoning AI model. If successful, this could mark a major milestone in the development of Artificial General Intelligence (AGI), where AI can think and reason like humans rather than just generating responses based on statistical probabilities.

4. OpenAI launches deep research for advanced research support

In early February, as excitement around DeepSeek remained high, OpenAI introduced Deep Research, a new feature in ChatGPT designed to help users conduct in-depth research more efficiently.

By aggregating information from multiple online sources, Deep Research generates detailed reports with full citations and displays the entire research process, including steps taken and sources used. This transparency allows users to track and verify their research workflow easily.

A key advantage of this feature is its ability to process various data formats—including text, images, and PDFs—providing a richer and more comprehensive research experience.

To use Deep Research, users simply enter a query and select the “Deep Research” icon before submitting their request. The process takes between 5 to 30 minutes, depending on the complexity of the question.

Currently, Deep Research is available to Pro, Plus, Team, Edu, and Enterprise users. Plus plan users can conduct 10 queries per month, Pro plan users are allowed 120 queries per month.

Due to high computational demands, OpenAI has stated that free-tier users will need to wait before accessing this feature.